Statistical Analysis Plan

to be addressed, namely: 1) feasibility in terms of safety, tolerance and adherence, 2)

preliminary efficacy as measured by load force symmetry from Baseline training Day 1 to Post training Day 5, and 3) preliminary efficacy of this intervention at improving stance time

symmetry from the mean of the Pre Test measures to Post Test. Other, additional measures of gait and balance were also collected and exploratory analyses conducted to examine the extent of interventional effects as well as correlative relationships between these parameters. Descriptive statistics (mean (SD)) are reported for age, gender, time since stroke event, side of impairment, MoCA, SIS, and FAC. All statistical analyses were completed using SPSS for Mac, Version 24 (IBM Corp©, SPSS Statistics, Chicago, Illinois). The G*Power version 3.1.9.3 software for Mac (Heinrich Heine Universitat Dusseldorf) was used for sample size calculations for future work.

3.5.1 Feasibility. Feasibility was evaluated in terms of safety, treatment tolerance and adherence.

Hypothesis 1: This intervention will be feasible in individuals post-stroke in terms of safety (as measured by severity and number of related adverse events), treatment tolerance (as measured by mean (SD) values of the maximum ratings of perceived exertion (RPE) ≤ 8), and adherence as measured by a mean adherence rate of 80%.

Safety is reported as a summary of the number and severity of adverse events. In terms of treatment tolerance, RPE and HR were recorded. RPE is reported descriptively using mean (SD) values of the group RPE by training day. Additionally, the percentage of age-predicted heart rate maximum (HRmax, 220-age) was calculated for each participant. Data is reported descriptively by group mean (SD) of HRmax by training day. A repeated measures ANOVA was used to determine change of both RPE and HRmax, respectively over the course of the training. Adherence rates were calculated as the number of sessions attended out of the total number of study sessions for each participant. The mean adherence for the group of participants is reported.

3.5.2 Efficacy. Efficacy of both load force symmetry and stance symmetry were evaluated as primary outcomes.

Hypothesis 2: This intervention will result in significantly improved load force symmetry from Baseline Day 1 to Post Training Day 5 in individuals post-stroke.

For this aim, symmetry ratios were calculated (paretic/non-paretic limb) from load force measures taken during gait on the treadmill recorded at the intervals previously described during Baseline Day 1 and Post Training Day 5. A mean of the Baseline (mBL) intervals (BL_time1, BL_time2, BL_time3) was calculated and used for statistical comparisons. Descriptive statistics (mean (SD)) are reported for mBL. Descriptive statistics (mean (SD)) are also reported for each of the Post Training intervals (PT_time1, PT_time2, PT_time3, PT_time4). Tests of normality were performed, and either a Wilcoxon Signed Ranks test or a paired t test was used to determine change after five treatment sessions by comparing mBL from Baseline Day 1 to Post Training Day 5 at each of the recorded time intervals, depending on normality of the data distribution. An alpha level was set at a p ≤ 0.05 level. Effect size and 95% confidence intervals were also calculated with SPSS. Effect sizes are reported in terms of partial eta squared and interpreted in accordance with guidelines (e.g. 0.00 to 0.20 is a small effect size, 0.40 to 0.60 is a moderate effect size, 0.80 or greater is a large effect size) (Richardson, 2011).

As an additional exploratory analysis of load force, within day changes of load force were also evaluated. Depending on normality of data distribution either a Wilcoxon Signed Ranks tests or paired t tests was used to compare same day mBL to each of the Post Training time intervals. Test statistics of these within day changes are reported with associated alpha levels, 95% confidence intervals, and effect sizes.

Hypothesis 3: This intervention will result in improved stance time symmetry from Pre Test to Post Test in individuals post-stroke.

This aim was addressed by first extracting the percentage of time spent in stance phase from inertial sensor data and calculating symmetry ratios (paretic/non-paretic limb). Once symmetry ratios were calculated, the mean was taken of the three Pre Test measures. This was done to minimize variability over the three days. The group mean of the Pre Test measures was used in a one-way repeated measures analysis of variance (ANOVA) to evaluate for changes between Pre Test to Post Test to Follow Up. Alpha level was set at p ≤ 0.05. If significant, pairwise comparisons were also examined to determine at which interval the change occurred. Bonferroni corrections were used to adjust for multiple comparisons. A 95% confidence interval and effect sizes were also calculated, and are interpreted as per recommended guidelines

(Richardson, 2011). Sample size calculations were made in preparation for future studies, and are also reported.

As an additional exploratory analysis of stance symmetry, a linear regression model was created using change scores from the mean of the three Pre Test measures to OG Pre Intervention Day 1, within day changes for each of the five training days (e.g. OG Pre Intervention to OG Post Intervention for each day of training), and from Post Test to Follow Up. Results of this regression model were used to determine the presence of a significant change over the course of the intervention. An additional regression model was used to examine changes between OG Pre Intervention measures on each day of training (e.g. OG Pre Intervention Day 1 to OG Pre

Post Intervention (e.g. OG Post Intervention Day 1 to OG Post Intervention Day 2, etc.). Slope and associated alpha levels of each of the three regression models are reported.

3.5.3 Exploratory Measures. The remainder of overground spatiotemporal gait parameters (e.g. time in double limb support, swing time symmetry, stride length, stride length symmetry, stride velocity, and gait velocity) were recorded as detailed above. Either a Wilcoxon Signed Ranks test or a one-way repeated measures ANOVA was completed to describe gains from Pre Test to Post Test to Follow Up. Alpha level was again set at a p ≤ 0.05 level. Effect sizes and 95% confidence intervals were calculated and reported. Similar to stance symmetry, three separate regression models were used to evaluate change over the course of the treatment for each variable (e.g. OG Pre Intervention to OG Pre Intervention, OG Post Intervention to OG Post Intervention, and Pre Test to Follow Up) and results of these are reported. Additionally, correlations were calculated between each of the exploratory spatiotemporal variables with stance symmetry. These are reported by Pearson correlation coefficients and are interpreted as per standard guidelines (e.g. |r| = 0.00 to 0.03 is a negligible correlation; 0.30 to 0.50 is a low correlation; 0.50 to 0.70 is a moderate correlation; 0.70 to 0.90 is a high correlation; and 0.90 to 1.00 is a very high correlation) (Mukaka, 2012).

Time spent in single limb support (SLS) was also collected in concert with other

spatiotemporal gait measures, and a repeated measures ANOVA performed from the mean of the Pre Test measures, to Post Test and to Follow up as a post hoc analysis. An alpha level was again set at p ≤ 0.05, and effect sizes were calculated with 95% confidence intervals. This additional analysis was performed to determine whether participants spent significantly more time on solely the affected limb during the gait cycle after completing the intervention, and to better understand the contribution of SLS to stance symmetry.

An independent rater, blinded to testing order, scored the BBS. Mean values of raw scores of the three Pre Test measures were taken to control for variability. Pre Test, Post Test and Follow Up BBS scores were analyzed using a one-way repeated measures ANOVA to evaluate for significant change between the three time points. An alpha level was again set at p ≤ 0.05. Effect sizes and 95% confidence intervals were also calculated and are reported.

All statistical assumptions (e.g. normality of distribution, homogeneity and

homoscedasticity) were checked prior to completing statistical analysis. For variables that were not normally distributed (e.g. in load force symmetry, Day 5 mBL, Days 4 and 5 PT_time1, swing symmetry and stride length symmetry were not normally distributed), a Wilcoxon Signed Ranks test was used rather than a paired t test or repeated measures ANOVA for comparisons.

Chapter IV

IV. Results - Subject Characteristics and Feasibility 4.1 Subject Characteristics

Of the 26 individuals contacted for participation, four elected not to participate either due to scheduling conflicts or long distance to travel, two reported other recent medical issues

interfering with ability to complete study requirements, and eight did not meet criteria for inclusion. Twelve individuals were enrolled for participation in the study. After reviewing all study procedures, risks and benefits of the study and answering questions, written informed consent was obtained. Of the 12 individuals enrolled, one subject withdrew due to unrelated nausea issues prior to initiating training and elected not to return, and 11 participants completed the study. Figure 4.1 presents the CONSORT flow diagram.

FIGURE 4.1 CONSORT flow diagram Screened for Eligibility

(n= 26) Enrollment (n= 12) Analyzed (n= 11) Excluded (n= 14)

• Not meeting inclusion criteria (n=8) • Declined to participate (n=4) • Other reasons (n=2) Withdrew participation (n =1) Pre Testing (n= 12) Training (n= 11) Post Testing (n= 11) Follow Up (n= 11)

Of the 11 participants that completed the study, the mean (SD) age was 50.18 (11.08) years, with age ranging between 24 to 63 years and a median age of 52 years. The mean (SD) number of months since stroke event was 71.24 (46.70), with a range between 9.30 and 128.87 months. Six of the 11 participants were male (five female). The left side was primarily affected for seven of the participants (four were right side affected). In the FAC scale, eight of the 11 participants were classified as a Category Five (independent navigating level and non-level surfaces), two were classified as a Category Four (independent navigating level surfaces only), and one participant as a Category Three (requires supervision level assist for navigating level surfaces). Nine of the participants scored ≥ 26 on the Montreal Cognitive Assessment (MoCA) indicating no cognitive impairment. One participant scored a 24 and another a 25 on the MoCA, indicative of mild cognitive impairment. As a measure of perceived quality of life, the mean (SD) SIS score of the 11 participants was 64.45 (10.70). Table 4.1 summarizes the participant characteristics at onset of study participation.

Table 4.1. Characteristics of study population.

Characteristics of each participant (n = 11) at onset of study participation. Mean (SD) also reported with data ranges.

Subject

ID Gender (in years) Months since CVA Age FAC (raw score) MoCA (raw score) SIS

01 M 60 128.87 5 26 62 02 F 63 90.60 5 29 73 03 M 24 126.43 5 29 77 05 F 52 114.00 5 29 66 06 M 61 98.33 5 30 75 07 F 57 17.50 5 29 75 08 M 45 9.30 4 24 46 09 F 52 38.67 4 25 68 10 F 50 100.93 5 30 63 11 M 46 16.20 3 26 55 12 M 42 42.77 5 26 49 Mean (SD) 6M / 5F (11.08) 50.18 Range 9.30 - 128.87 71.24 (46.70) 8 = Cat 5 2 = Cat 4 1 = Cat 3 27.55 (2.16) Range 24-30 Range 46 - 77 64.45 (10.70) 4.2 Feasibility

Feasibility is reported in terms of safety, treatment tolerance and adherence.

4.2.1 Safety. No adverse events or serious adverse events were experienced during the course of this study.

4.2.2 Treatment Tolerance. The mean (SD) of RPE taken throughout daily training sessions was used as the primary measure of treatment tolerance. The group mean (SD) RPE over the five-day intervention was 3.61 (0.23). The group mean (SD) by training day is reported in Table 4.2. A repeated measures ANOVA showed no difference in RPE between days of training (p > 0.05). As an additional measure of treatment tolerance, the group mean (SD) of the percentage of age-predicted HRmax over the course of the five-day training was 55.96 (1.46).

Similar to RPE, there was no difference in the percentage of HRmax over the course of training (p >0.05). The group means (SD) for each day of training are reported in Table 4.3.

Table 4.2 Group means (SD) with daily mean ranges of the RPE reported by training day over the course of the intervention.

Training

Day Mean (SD) RPE Range

Day 1 3.76 (1.39) 2.38 - 6.38 Day 2 3.96 (1.27) 1.63 - 6.14 Day 3 3.56 (1.50) 1.63 - 6.50 Day 4 3.63 (1.06) 2.00 - 5.00 Day 5 3.16 (0.95) 1.38 - 4.38 Mean 3.61 (0.23)

Table 4.3 Group means (SD) with daily mean ranges of the age-predicted percentage of HRmax reported by training day over the course of the intervention.

Training Day Mean (SD) Percentage of Age-Predicted HRmax Range Day 1 54.72 (10.87) 41.47 - 73.47 Day 2 58.21 (13.39) 38.32 - 84.71 Day 3 57.23 (11.81) 42.35 - 76.43 Day 4 55.33 (9.96) 39.56 - 75.31 Day 5 54.32 (9.87) 39.41- 70.06 Mean 55.96 (1.46)

4.2.3 Adherence. There were a total of nine study visits to be completed by each participant, three Pre Test assessments, five days of training and one Follow Up visit. Of the eleven participants who completed the study, nine completed all study visits (100% adherence) two missed a single day of training (80% adherence), yielding an overall adherence rate of 96.4%.

Chapter V

Results – Efficacy

V. Results - Efficacy

5.1.1 Load Force Symmetry. On normality testing, Days 4 and 5 early Post Training (PT_time1) and Day 5 Baseline (mBL) data were not normally distributed; therefore, a Wilcoxon Signed Ranks test was used in lieu of a paired t test for analyses involving these data sets. All other data were normally distributed. The mean (SD) Baseline (mBL) load force symmetry for all intervals recorded during the Baseline period on the treadmill during Day 1 of training was 0.71 (0.09). Results from treadmill recordings on the completion of five treatment sessions yielded a mean (SD) during early Post Training (PT_time1) of 0.83 (0.23), to 0.82 (0.19) at PT_time2, to 0.79 (0.14) at PT_time3 and to 0.81 (0.19) at late Post Training (PT_time4) on Day 5. On statistical comparison, the mean change from Day 1 mBL to Day 5 PT_time1 was not statistically significant (p >0.05). Confidence intervals and effect sizes are reported in Table 5.1. Comparisons of Day 1 mBL values to subsequent Day 5 Post Training intervals (PT_time2, PT_time3, and PT_time4) likewise, were not statistically significant (p>0.05). See Table 5.1 for a list of each of the comparisons (Day 1 mBL to each of the Post Training intervals) with

corresponding change in load force symmetry ratio, t/Z statistic, p values, 95% confidence intervals and effect sizes. Additionally, Figure 5.1 graphically represents the change from the mBL on Day 1 through the Post Training intervals on Day 5.

Table 5.1. Load force symmetry at Day 1 mean Baseline to Day 5 at each of the Post Training intervals.

Mean (SD) values for Day 1 mean Baseline (mBL) and each of the Post Training intervals (PT_time1, PT_time2, PT_time3, PT_time4) on Day 5. Comparisons were made between the mBL and each of the PT intervals and are also included with corresponding t/Z statistic, p values, effect estimates, 95% confidence intervals and effect size.

Force Symmetry

Ratio

Mean (SD) t/Z Value p value

Mean Effect

Estimate 95% CI Partial Eta Sq

Day1 mBL 0.71 (0.09)

Day5 PT_time1 0.83 (0.23) Z = 1.778 p = 0.075 0.124 (-0.027, 0.276) 0.251

Day5 PT_time2 0.82 (0.19) t = 2.134 p = 0.059 0.112 (-0.005, 0.229) 0.313

Day5 PT_time3 0.79 (0.14) t = 2.157 p = 0.056 0.087 (-0.003, 0.177) 0.318

Day5 PT_time4 0.81 (0.19) t = 2.032 p = 0.070 0.103 (-0.010, 0.217) 0.292

FIGURE 5.1. Load force symmetry ratios from Day 1 mean of Baseline (mBL) values to the completion of Post Training (PT_time1, PT_time2, PT_time3, PT_time4) on Day 5.

Results of within day comparisons on each of the training days demonstrated a statistically significant reduction in symmetry on Day 3 (-0.03) of training, (t = -2.757, p= 0.025), with an effect size of 0.487, and a 95% confidence interval ranging (-0.056, -0.005). No significant change in symmetry was achieved on any other day of training (p >0.05). Table 5.2 provides detailed mean (SD) values for each of the intervals measured during Baseline and Post Training on each day of training. Associated t/Z statistic, p values, 95% confidence intervals and effect sizes are also included.

Table 5.2. Mean (SD) load force symmetry on each day of training and within day changes. Values for mean Baseline (mBL) and at each time interval for Post Training (PT_time1,

PT_time2, PT_time3, and PT_time4) on each day of training. Mean change between mBL and PT_time1 is also included. Associated t/Z statistic for the change between mBL to PT_time1 are included with respective p values, effect estimates, 95% confidence intervals and effect sizes.

* denotes statistical significance at p < 0.05

Day 1 Day 2 Day 3 Day 4 Day 5 Mean

mBL 0.71 (0.09) 0.73 (0.09) 0.79 (0.10) 0.77 (0.10) 0.79 (0.19) 0.76 (0.04) PT_time1 0.72 (0.07) 0.75 (0.10) 0.75 (0.09) 0.82 (0.22) 0.83 (0.23) 0.77 (0.05) PT_time2 0.73 (0.10) 0.75 (0.11) 0.78 (0.09) 0.83 (0.23) 0.82 (0.19) 0.78 (0.04) PT_time3 0.71 (0.07) 0.75 (0.10) 0.79 (0.10) 0.80 (0.17) 0.79 (0.14) 0.77 (0.04) PT_time4 0.70 (0.10) 0.74 (0.09) 0.77 (0.08) 0.84 (0.21) 0.81 (0.19) 0.77 (0.05) Change mBL to PT_time1 0.01 0.02 -0.03 0.05 0.04 0.02 t/Z Value t = 0.692 t = 0.991 t = - 2.757 Z = - 0.800 Z = - 0.153 p value p = 0.505 p = 0.345 *p = 0.025 p = 0.424 p = 0.878 Mean Effect Estimate 0.013 0.017 -0.031 0.052 0.037 95% CI (-0.029, 0.054) (-0.021, 0.055) (-0.056, -0.005) (-0.054, 0.158) (-0.037, 0.111) Partial Eta Sq 0.046 0.089 0.487 0.107 0.110

5.2 Overground Measures

5.2.1 Stance Symmetry. Statistical assumptions were met, and results of a one-way repeated measures ANOVA demonstrated a significant main effect (F=8.498, p = 0.002) between mean Pre Test measures recorded during week one, prior to initiating the intervention to Post Test on completion of the intervention, and to the one-week Follow Up, with a 95% confidence interval ranging (0.005, 0.096) and an effect size of 0.459. A Bonferonni correction was used, and pairwise comparisons revealed a significant effect between the mean of the three Pre Test measures to Post Test (p = 0.028), but no statistically significant difference between Pre Test and Follow Up. A group mean (SD) gain of a 0.05 (0.05) was demonstrated in stance symmetry from Pre Test to Post Test; however, this gain was reduced to a net 0.02 gain from Pre Test to Follow Up.

Results of a regression analysis showed no significant change between daily measures recorded prior to training (e.g. OG Pre Intervention Day 1 to OG Pre Intervention Day 2, etc.), or for measures recorded after daily training (e.g. OG Post Intervention Day 1 to OG Post

Intervention Day 2, etc.) (p >0.05). In contrast, there was a significant effect of training noted from within day changes (e.g. OG Pre Intervention Day 1 to OG Post Intervention Day 1, etc.) over the time course of the intervention (p = 0.027), with a positive slope of 0.002 between time points. See Figure 5.2 for a graphical representation of stance symmetry over the course of the intervention. See Table 5.4 for a summary of mean (SD) values at time points used for statistical analysis with corresponding output.

FIGURE 5.2. Daily mean stance symmetry at each day of the intervention.

Points represent the mean of the three days of Pre Test, to OG Pre/Post Intervention each day, and Follow Up.

Sample size calculations for future studies were made using mean (SD) and effect size values from Pre Test to Post Test of overground stance symmetry. A total of 14 participants would be required to demonstrate change at a level of 80% statistical power.

5.2.2. Exploratory Measures. Correlations were calculated between stance symmetry and each of the other spatiotemporal gait parameters. Significant positive correlations were noted between stance and swing symmetry, stance symmetry and stride length, and stance symmetry and stride velocity. Significant negative correlations were demonstrated between stance symmetry and double limb support and between stance symmetry and gait velocity. See Table 5.3 for each of the gait parameters correlated with stance symmetry and the corresponding correlation

coefficients. 0.75 0.8 0.85 0.9 0.95 1 Mean_ PreT est D1_O G Pre D1_O G Post D2_O G Pre D2_O G Post D3_O G Pre D3_O G Post D4_O G Pre D4_O G Post D5_O G Pre Post T est Follo w Up *p = 0.027, slope = 0.002

*

Table 5.3. Correlations between stance symmetry and other spatiotemporal gait parameters.

Variables Tested r p value

Swing Symmetry 0.894 p = 0.01

Double Limb Support -0.314 p = 0.01

Stride Length 0.472 p = 0.01

Stride Velocity 0.492 p = 0.01

Gait Velocity -0.446 p = 0.01

On tests of normality, swing symmetry and stride length symmetry were not normally distributed, so a Wilcoxon Signed Ranks test was used. There was a significant difference in swing symmetry between mean Pre Test to Post Test (Z = -2.223, p = 0.026) and Post Test and Follow Up (Z = -2.490, p = 0.013), but not between mean Pre Test and Follow Up (p >0.05). Confidence intervals and effect sizes are reported in Table 5.4. Similar to stance symmetry, on regression analysis, no significant changes were noted between time points recorded prior to daily intervention, nor on time points recorded after daily intervention; however, significant changes were noted within day of treatment over the course of the intervention (p = 0.023, slope = 0.004). See Figure 5.3. No significant changes were noted between Pre Test, Post Test, and Follow up for stride length symmetry.

FIGURE 5.3. Daily mean swing symmetry at each day of the intervention.

Points represent the mean of the three days of Pre Test, to OG Pre/Post Intervention each day, and Follow Up.

Other spatiotemporal gait parameters including double limb support, stride length, stride velocity, and gait velocity demonstrated a normal distribution and other statistical assumptions were met, therefore a one-way repeated measures ANOVA was performed. Results showed no significant difference between values of the mean of the Pre Test measures to Post Test to Follow Up for any of the other variables. Similarly, regression analyses showed no significant changes between the time points measured prior to daily intervention, after daily intervention, nor within daily interventions for any other spatiotemporal variables.

On post hoc analysis of single limb support, no significant change was noted from the mean of the Pre Test measures to Post Test to Follow Up (p > 0.05). A 95% confidence interval ranged (-0.991,2.204) with an effect size of 0.071. See Table 5.4 for group means (SD) at Pre Test, Post Test and Follow Up.

Correlations between stance symmetry and the Berg Balance Scale demonstrated a positive correlation (r = 0.401, p = 0.05). Results of a one-way, repeated measures ANOVA of the Berg Balance Scale showed no significant main effect for change over the course of the

0.5 0.55 0.6 0.65 0.7 0.75 Mean_ PreT est D1_O G Pre D1_O G Post D2_O G Pre D2_O G Post D3_O G Pre D3_O G Post D4_O G Pre D4_O G Post D5_O G Pre Post T est Follo w Up *p = 0.023, slope = 0.004

*

intervention for balance (p > 0.05). All primary and additional exploratory spatiotemporal gait parameters are listed in Table 5.4 with associated means (SD), F/Z statistic, p values, confidence interval values and effect sizes. See Table 5.4.

Table 5.4. Spatiotemporal gait parameters with F/Z Statistic, confidence intervals and effect sizes.

Values at Pre Test, Post Test and Follow Up represent group means (SD). Pre Test values